广义概率Tsallis熵的快速多阈值图像分割

Fast Multilevel Thresholding for Image Segmentation Based on Generalized Probability Tsallis Entropy

针对传统的熵多阈值法存在的计算复杂度高和分割不准确等问题,提出了一种基于广义概率Tsallis熵的快速多阈值图像分割方法。首先,对传统的灰度概率进行修改得到广义概率以构建广义概率Tsallis熵。然后,通过直方图均值自动确定Tsallis熵参数以解决参数不易选择的问题。随后,将GPTE正确拓展到多阈值分割方法中使得分割更准确。最后,将差分进化(Differential evolution,DE)算法与递推算法有机结合应用于GPTE多阈值法中以解决计算复杂度高的问题。图像分割实验结果表明,与基于传统的熵多阈值法相比,本文提出的方法不仅分割更准确,自适应性更强,而且运行速度更快。

To overcome the shortcomings of entropy-based multilevel segmentation methods, such as high computational complexity and poor segmentation performance, a fast multilevel thresholding for image segmentation method based on generalized probability Tsallis entropy is proposed. First the traditional gray probability is modified into generalized probability to form generalized probability Tsallis entropy （GPTE）. Then the parameters of GPTE are chosen adaptively through the image histogram average val- ue. Thirdly, a multilevel thresholding method based on GPTE is formulated to get more effective seg- mentation. Finally the differential evolution（DE） and the recursive algorithm are combined and intro- duced into the multilevel thresholding method to find the best threshold vector quickly. Experimental re- sults of image segmentation show that the propos method can obtain better segmentation results and a- daptability with less computation time compared with the traditional entropy-based multilevel segmenta- tion methods.